US 7190733 B2 Abstract An algorithm for correcting the output of an analog I/Q demodulator without the need for calibration or storing state information. The output of the analog I/Q demodulator is digitized, and the discrete-time samples are divided into segments. A digital frequency transform (e.g., a Discrete Fourier Transform) is computed for each segment. The effects of the non-ideal I/Q demodulator are removed by identifying a set of image frequency candidates for each digital frequency transform, and for each image frequency candidate, estimating a set of demodulator imbalance parameters to characterize the demodulator imbalance at that frequency and correcting the digital frequency transform at the image frequency candidate using the imbalance parameters in order to minimize the effects of the imbalance. Each digital frequency transform is corrected independently and consequently no persistent state information needs to be saved between transforms.
Claims(18) 1. A method for correcting the output of an analog I/Q demodulator, comprising:
a. computing a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator;
b. identifying a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency;
c. for each image frequency candidate, estimating I/Q demodulator imbalance parameters; and
d. correcting the value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters.
2. The method of
3. The method of
4. The method of
5. The method of
_{m}=γ(Y_{m}−βY_{−m}*), where Y_{m }is an uncorrected value of the frequency transform at the image frequency candidate m and Y′_{m }is the corrected value of the digital frequency transform at the image frequency candidate m, and where the set of I/Q demodulator imbalance parameters comprise γ and β.6. The method of
_{m }as an average of the values of the digital frequency transform at neighboring image frequency candidates.7. The method of
8. A method for correcting the output of an analog I/Q demodulator, comprising:
a. computing a digital frequency transform for each sample of a time-based output of the analog demodulator;
b. analyzing the digital frequency transform to identify image frequency candidates; and
c. removing an image frequency candidate if it actually is an image, otherwise, not removing the image frequency candidate.
9. A processor readable medium storing instructions that, when executed by a processor, cause the processor to:
a. compute a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator;
b. identify a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency produced by the analog I/Q demodulator;
c. for each image frequency candidate, estimate I/Q demodulator imbalance parameters; and
d. correct a value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters.
10. The processor readable medium of
11. The processor readable medium of
12. The processor readable medium of
13. The processor readable medium of
_{m}=γ(Y_{m}−βY_{−m}*), where Y_{m }is an uncorrected value of the frequency transform at the image frequency candidate m and Y′_{m }is the corrected value of the digital frequency transform at the image frequency candidate m, and where the set of I/Q demodulator imbalance parameters comprise γ and β.14. The processor readable medium of
_{m }as an average of the values of the digital frequency transform at neighboring image frequency candidates if a value of the digital frequency transform at the image frequency candidate appears as an image of a relatively narrow-band signal.15. The processor readable medium of
16. A radio receiver system, comprising:
a. a radio receiver that receives radio frequency signals and includes an I/Q analog demodulator that demodulates the radio frequency signals to produce an analog baseband signal;
b. an analog-to-digital converter coupled to the radio receiver that converts the analog baseband signal to a digital signal; and
c. a processor coupled to the analog-to-digital converter that is programmable to:
i. compute a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator;
ii. identify a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency produced by the analog I/Q demodulator;
iii. for each image frequency candidate, estimate I/Q demodulator imbalance parameters; and
iv. correct a value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters.
17. A method for correcting the output of an analog I/Q demodulator, comprising:
a. computing a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator;
b. identifying a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency;
c. for each image frequency candidate, computing I/Q demodulator imbalance parameters such that the digital frequency transform is conjugate symmetric at the image frequency candidate; and
d. correcting a value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters, by computing computing Y′
_{m}=γ(Y_{m}−βY_{−m}*), where Y_{m }is an uncorrected value of the frequency transform at the image frequency candidate m and Y′_{m }is the corrected value of the digital frequency transform at the image frequency candidate m, and where the set of I/Q demodulator imbalance parameters comprise γ and β.18. A method for correcting the output of analog I/Q demodulator, comprising:
a. converting the output of the analog I/Q demodulator to segments of discrete-time samples;
b. computing a digital frequency transform for each segment of discrete-time samples;
c. identifying a set of image frequency candidates for each digital frequency transform;
d. for each image frequency candidate, estimating a set of demodulator imbalance parameters to characterize the demodulator imbalance at that frequency; and
e. correcting the digital frequency transform at the image frequency candidate using the imbalance parameters in order to minimize the effects of the imbalance.
Description This application claims priority to U.S. Provisional Application No. 60/629,917, filed Nov. 22, 2004, the entirety of which is incorporated herein by reference. The present invention relates to radio receiver devices, and more particularly to techniques for correcting for inaccuracies in in-phase/quadrature-phase (I/Q) demodulators. As shown in State-of-the-art I/Q demodulator implementations typically provide an image suppression of 35–40 dB, which is sufficient to guarantee acceptable performance in most applications. However, for real-time spectrum management applications where it is desired to detect the types of signals occurring in a shared frequency band, the required image suppression is much larger than 35 dB. In fact, in some cases an image suppression of approximately 90 dB is required. Image artifacts above the noise floor make spectrum analysis displays confusing. In addition, signal classification and pulse detection functions can be confused by the image artifacts if they are not properly removed. There are several known techniques for mitigating the image artifacts inherently created by non-ideal I/Q demodulators. According to one technique, an intermediate frequency (IF)-based receiver is used and digital I/Q demodulation is performed at IF. The drawback of this technique is that IF receivers can only be implemented using discrete analog circuits. Another technique involves using calibration to measure the amplitude and phase imbalance of the I/Q modulator and correct for the imbalances digitally. The problems with this technique are that it can only provide up to 65 dB of suppression over temperature, and a relatively long calibration time is required at manufacture. Still another approach is to use a complex very low intermediate frequency (VLIF)-based receiver. With this approach, however, out-of-band signals can appear as if they were in-band, still requiring calibration to minimize this effect. An improved approach for removing the effects of I/Q demodulator imbalance is needed that is efficient and flexible to implement, and capable of providing significant image suppression required for certain communication applications, such as spectrum analysis or management applications. Briefly, the output of an analog I/Q demodulator is digitized, the discrete-time samples are divided into segments, and a digital frequency transform (e.g., a Discrete Fourier Transform) is computed for each segment. The effects of the non-ideal I/Q demodulator are removed by identifying a set of image frequency candidates for each digital frequency transform, and for each image frequency candidate, estimating a set of demodulator imbalance parameters to characterize the demodulator imbalance at that frequency and correcting the digital frequency transform at the image frequency candidate using the imbalance parameters in order to minimize the effects of the imbalance. In this algorithm, no calibration is required. Each digital frequency transform is corrected independently and consequently no persistent state information needs to be saved between transforms. The image artifacts are attenuated into the noise floor of each input transform such that the amount of image suppression is always sufficient. The above and other advantages of this technique will become more apparent when reference is made to the following description taken in conjunction with the accompanying drawings. An I/Q modulated RF signal has the form.
A discrete-time version of equation (1) can be written as:
Applying a window function W
Taking N-point digital Fourier transforms (DFTs) of both sides of equation (3) yields
Assuming s then, S
The image suppression at the demodulator output (in dB) is
An arbitrary finite-length signal has a unique DFT, and can therefore be written as a sum of sinusoids:
In this case, equation (4) becomes
Since the DFT window spectrum W
Both sides of equation (6) can be conjugated to solve for S
The spectrum correction algorithm is motivated from the observation that in equations (6) and (7), the spectrum Y Y _{k} =β·Y _{−k}* for some βTurning to In addition, the output for the spectrum correction module The DFT module With reference to The effects of the non-ideal I/Q demodulator are removed by identifying a set of image frequency candidates for each digital frequency transform. For each image frequency candidate, a set of demodulator imbalance parameters are estimated to characterize the demodulator imbalance at that frequency and used to correct the digital frequency transform at the image frequency candidate in order to minimize the effects of the imbalance. More specifically, in step Once steps The set M of image frequency candidates M are found in step The cost function used in step -
- 1. Find the multiplier β to best make Y “appear conjugate symmetric” at the image frequencies.
- 2. Account for frequency dependence in β because in practice, the amplitude and phase offset of a typical analog I/Q demodulator varies in-band over frequency due to (a) amplitude and group delay skew in the I and Q lowpass filters, and (generally to a lesser extent) (b) amplitude and phase variation in the cos( ) and sin( ) function generators feeding the mixers. For certain spectrum analysis applications, it has been determined that the offset remains constant over roughly 2 MHz of the 20 MHz output bandwidth.
The cost function is defined mathematically as follows:
In other words, the sum in equation (10) is carried out over all candidates in M that are within 2 MHz of the candidate m at which spectrum correction is being applied. By taking derivatives and equating to zero, the minimizer of equation (10) can be shown to be:
To summarize, the spectrum correction algorithm described above involves, on a per-digital frequency transform basis, finding a set of image frequency candidates m, and for each image frequency candidate, estimating the I/Q demodulator imbalance parameters that make the digital frequency transform look as it would if it came from an ideal analog demodulator. That is, at each image frequency candidate, the algorithm estimates a set of I/Q demodulator imbalance parameters (β and γ), and then correction is applied at the image frequency candidate m using equation (8B). Thus, the estimated I/Q demodulator imbalance parameters are frequency dependent, or more specifically image frequency candidate dependent. The I/Q demodulator imbalance parameters are computed using the cost function of equation (10) such that when applied to an image frequency candidate using equation (8B), it either removes the image frequency candidate if it actually is an image and leaves the digital frequency transform otherwise unaffected, or it leaves it unaltered if it is not an image. Thus, if the step of identifying an image frequency candidate wrongly flags a frequency of the DFT as being an image candidate, e.g., there really is not an image artifact at frequency m, but rather just another signal that happens to be 25 or more dB lower than the signal at the negative frequency, the algorithm will not alter the DFT data at that frequency. Suppose an I/Q demodulator has an amplitude and phase imbalance characterized by which α is estimated using the spectrum correction algorithm described above. The following explains how much image suppression the algorithm should deliver. If m is an image frequency, then from equation (6)
and the image suppression of the uncorrected I/Q demodulator is
Substituting the above formulas for Y Examples of the performance of the spectrum correction algorithm are shown in Signal Thus, in The following observations were noted from the simulations for various SNRs, FFT sizes and input signals. -
- (1) Images are suppressed only enough to put them into the noise floor.
- (2) Infinite image suppression is obtained as the input SNR goes to infinity.
- (3) When the input signal is a tone, a small notch appears in the max-hold plot at the image frequencies. This is because the sum in equation (11) only has only a small number of terms carrying signal information, causing the estimator to suppress both image and noise, rather than just the image. This effect can be corrected by replacing the image corrected samples with interpolated samples on either side of the image.
In sum, a method is provided for correcting the output of an analog I/Q demodulator, comprising computing a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator; identifying a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency; for each image frequency candidate, estimating I/Q demodulator imbalance parameters; and correcting the value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters. Said another way, a method is provided for correcting the output of an analog I/Q demodulator, comprising computing a digital frequency transform for each sample of a time-based output of the analog demodulator; analyzing the digital frequency transform to identify image frequency candidates; and removing an image frequency candidate if it actually is an image, otherwise, not removing the image frequency candidate. Said still another way, a method is provided for correcting the output of analog I/Q demodulator, comprising: converting the output of the analog I/Q demodulator to segments of discrete-time samples; computing a digital frequency transform for each segment of discrete-time samples; identifying a set of image frequency candidates for each digital frequency transform; for each image frequency candidate, estimating a set of demodulator imbalance parameters to characterize the demodulator imbalance at that frequency; and correcting the digital frequency transform at the image frequency candidate using the imbalance parameters in order to minimize the effects of the imbalance. Said yet another way, a method is provided for correcting the output of an analog I/Q demodulator, comprising: computing a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator; identifying a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency; for each image frequency candidate, computing I/Q demodulator imbalance parameters such that the digital frequency transform is conjugate symmetric at the image frequency candidate; and correcting a value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters, by computing Y′ The techniques described herein may be embodied by instructions stored on a processor readable medium, including instructions that, when executed by a processor, cause the processor to: compute a time-based sequence of digital frequency transforms for the time-domain output of the I/Q analog demodulator; identify a set of one or more image frequency candidates in the digital frequency transform, wherein at each image frequency candidate there may be a signal that is an image of the input signal to the analog I/Q demodulator at the negative frequency produced by the analog I/Q demodulator; for each image frequency candidate, estimate I/Q demodulator imbalance parameters; and correct a value of the digital frequency transform at the image frequency candidate using the set of I/Q demodulator imbalance parameters. The processor so programmed may be part of a radio receiver system that also includes a radio receiver that receives radio frequency signals and includes an I/Q analog demodulator that demodulates the radio frequency signals to produce an analog baseband signal, and an analog-to-digital converter coupled to the radio receiver that converts the analog baseband signal to a digital signal. The above description is intended by way of example only. Patent Citations
Non-Patent Citations
Referenced by
Classifications
Legal Events
Rotate |